Sounding Names / Religion

Overview

Sounding Names predicts religion and race-ethnicity attributes from personal names using two linked inference systems: a character n-gram TF-IDF plus SVM religion classifier and a CNN-based race-ethnicity classifier.

Data and Sources

• Name datasets in CSV format with optional household-name concat fields for religion tasks.
• Pretrained vectorizers, tokenizers, encoders, and model weights bundled in-repo.
• Optional GIS/demographic prior inputs for enhanced race-ethnicity model modes.

Workflow

• Normalize and clean names with transliteration and character-level preprocessing.
• Run binary or multiclass religion inference depending on selected model artifacts.
• Run CNN race-ethnicity inference in text-only or prior-augmented mode.

Outputs

• Prediction files with class labels plus score distributions for audit interpretation.
• Notebook and script pathways for local or Colab execution.
• Operational documentation for environment compatibility and model constraints.